Comparative Investigation of the TIMSS 2011 Mathematics Subtest's Item Parameters According to Classical Test Theory And Item Response Theory With BILOG MG, MULTILOG and R Programs

FAZİLET GÜL İNCE

Comparative Investigation of the TIMSS 2011 Mathematics Subtest's Item Parameters According to Classical Test Theory And Item Response Theory With BILOG MG, MULTILOG and R Programs

Thesis Type: Postgraduate

Institution Of The Thesis: Gazi University, Eğitim Bilimleri Enstitüsü, EĞİTİMDE ÖLÇME VE DEĞERLENDİRME ANABİLİM DALI, Turkey

Approval Date: 2016

Thesis Language: Turkish

Student: FAZİLET GÜL İNCE

Supervisor: MEHTAP ÇAKAN

Open Archive Collection: AVESIS Open Access Collection

Abstract:

The purpose of this study is to scrutinize the item parameters and characteristics of 4th grade mathematics subtest in TIMSS 2011, Turkey, according to Classical Test Theory (CTT) and one parameter, two parameter and three parameter logistic models of Item Response Theory (IRT) by using BILOG, MG, R, MULTILOG programs. The sample of the research consists of 543 students who participated in TIMSS 2011, Turkey and took a subtest 11th which from 4th grade mathematics tests. This research is a descriptive study of background study as the existing institutions were examined using real data, analyzed and interpreted. According to research results, it was found out that the correlation between difficulty coefficients of dichotomous items of CTT and b parameters of IRT was found negative, high and significant. In addition, while correlation between discrimination coefficients of CTT and a parameters of IRT for one parameter logistic and two parameter logistic models was positive, high and significant, for 3 parameter logistic model, it was positive and significant in the middle level. When the results obtained using BILOG, MG, MULTILOG and R were investigated, it was found out that BILOG MG and MULTILOG programs yielded almost similar results. When model-data fit was scrutinized, it was deduced that while the fittest model for BILOG was 2 parameter logistic model, according to R and MULTILOG 3 parameter logistic model was the fittest one. For reliability analysis, Cronbach alfa internal consistency coefficient was calculated using CTT, and marginal reliability coefficients were calculated using IRT. The difference between reliability coefficients from CTT and IRT was tested and for three software in 3 parameter model, and in 2 parameter model, difference between reliability coefficient estimated by BILOG and Cronbach alpha coefficient was significant. Finally, the standard error of the programs were calculated and it was found that there was a significant difference for b parameter of 1PL model. In general, it was detected that R, BILOG and MG programs yielded lower error than MULTILOG program.